Parallelization of artificial neural network training algorithms: A financial forecasting application

Artificial neural networks (ANN) are widely used to solve series prediction problems such as prices of financial instruments. Backpropagation is the most common artificial neural training algorithm. This paper discusses results obtained with the parallelization of the backpropagation algorithm used to train a network that forecasts the S&P500 Index. Training this ANN involves the processing of vast amounts of historical financial data which is time consuming. Financial markets; however, constitute fast paced environments where decisions need to make shortly after new information becomes available. Parallelizing the backpropagation algorithm to run on four processors simultaneously resulted in a reduction of 61% in training time compared to the same algorithm running without parallelization.